Oracle Select AI: The Architecture of Natural Language Data

In today’s data-driven world, the ability to ask questions and get immediate answers from your database is a game-changer. Oracle Select AI is a revolutionary developer assistant that allows both beginners and experts to interact with private enterprise data using natural language. By bridging the gap between Large Language Models (LLMs) and your specific database, Select AI makes it possible to derive value from both structured tables and unstructured text.

Oracle Select AI, by utilizing prompt augmentation, converts plain English into SQL queries without exposing sensitive table data to external models. The technology supports various LLM providers and incorporates advanced features like Retrieval Augmented Generation (RAG) for unstructured documents and synthetic data generation for testing. To ensure enterprise-grade security, the platform maintains strict access controls and offers private network configurations. 

👉 What is Oracle Select AI

👉  Why Metadata is Important for Select AI

How Select AI Works: The Magic of Prompt Augmentation

 The secret to Select AI's accuracy is a process called Prompt Augmentation.

Prompt augmentation bridges the gap between natural language and SQL by enriching a user's plain English question with specific database metadata before it is processed by a Large Language Model (LLM). This process ensures the AI has the necessary context to understand a unique database schema and generate accurate, executable code.\

 When you ask a question in plain English, Select AI does not simply send that question to an AI provider. Instead, it follows a structured workflow:

Phase 1: Metadata Integration: Select AI automatically adds database metadata—such as table names, column names, data types, and comments—to your prompt.

Phase 2: Prompt Construction: This "augmented prompt" provides the LLM with the context it needs to understand your unique schema.

 

Phase 3: SQL Generation: The LLM uses its general reasoning to generate a syntactically correct SQL query specific to your database.

 4.     Phase 4: Execution: For most actions, the generated SQL is executed locally on your database, ensuring that actual table data is never sent to the LLM during the query generation phase.

 The Core Components of Select AI

 

To deliver this seamless experience, Select AI relies on several tightly integrated components:

·         AI Profiles: These are predefined configurations created using the dbms_cloud_ai package. A profile specifies the AI provider (like OCI Gen AI or OpenAI), the specific model to use (e.g., GPT-4o), and the Object List of tables the AI is allowed to see.

LLM Providers: Select AI is flexible, supporting a wide range of providers including OCI Generative AI, OpenAI, Azure, Cohere, Google Gemini, and Anthropic.

AI Vector Search: For handling unstructured data, Select AI integrates with 23ai AI Vector Search to perform semantic similarity searches based on the "meaning" of the data rather than just keywords.

 Interactive Actions: Users can choose how they interact with the AI via specific keywords:

o   run_sql: The default action that generates and executes a query.

o   show_sql: Displays the generated code for review.

o   explain_sql: Provides a step-by-step natural language breakdown of what the SQL query is doing.

o   narrate: Sends the query results back to the LLM to generate a conversational summary of the data.

 Beyond SQL: Advanced Capabilities

Select AI isn't just for querying tables. It includes advanced features that address modern enterprise needs:

Retrieval Augmented Generation (RAG) RAG allows you to give an LLM new, private knowledge from documents like PDFs or blog posts without needing to "fine-tune" the model. Select AI automates the creation of vector indexes and the retrieval of relevant text chunks to provide authoritative answers based on your internal knowledge base.

Synthetic Data Generation For developers who need realistic test data without exposing sensitive production information, Select AI can generate synthetic records. By looking at your existing schema and sample rows, it creates new, realistic data that maintains referential integrity.

 

Security and Privacy First

Security is built into the architecture. Select AI enforces all existing Oracle security policies and data access rules for every query. While most actions only share metadata, it is important to note that the narrate action is the exception—it sends the actual rowset to the LLM to create a summary. For highly sensitive environments, Oracle offers the ability to use private endpoints or privately hosted models to ensure data never leaves your controlled network.